git.oblomov.eu Git - git/blob - builtin/pack-objects.c

   1 #include "builtin.h"
   2 #include "cache.h"
   3 #include "attr.h"
   4 #include "object.h"
   5 #include "blob.h"
   6 #include "commit.h"
   7 #include "tag.h"
   8 #include "tree.h"
   9 #include "delta.h"
  10 #include "pack.h"
  11 #include "pack-revindex.h"
  12 #include "csum-file.h"
  13 #include "tree-walk.h"
  14 #include "diff.h"
  15 #include "revision.h"
  16 #include "list-objects.h"
  17 #include "pack-objects.h"
  18 #include "progress.h"
  19 #include "refs.h"
  20 #include "streaming.h"
  21 #include "thread-utils.h"
  22 #include "pack-bitmap.h"
  23 #include "reachable.h"
  24 #include "sha1-array.h"
  25 #include "argv-array.h"
  26
  27 static const char *pack_usage[] = {
  28         N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
  29         N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
  30         NULL
  31 };
  32
  33 /*
  34  * Objects we are going to pack are collected in the `to_pack` structure.
  35  * It contains an array (dynamically expanded) of the object data, and a map
  36  * that can resolve SHA1s to their position in the array.
  37  */
  38 static struct packing_data to_pack;
  39
  40 static struct pack_idx_entry **written_list;
  41 static uint32_t nr_result, nr_written;
  42
  43 static int non_empty;
  44 static int reuse_delta = 1, reuse_object = 1;
  45 static int keep_unreachable, unpack_unreachable, include_tag;
  46 static unsigned long unpack_unreachable_expiration;
  47 static int pack_loose_unreachable;
  48 static int local;
  49 static int have_non_local_packs;
  50 static int incremental;
  51 static int ignore_packed_keep;
  52 static int allow_ofs_delta;
  53 static struct pack_idx_option pack_idx_opts;
  54 static const char *base_name;
  55 static int progress = 1;
  56 static int window = 10;
  57 static unsigned long pack_size_limit;
  58 static int depth = 50;
  59 static int delta_search_threads;
  60 static int pack_to_stdout;
  61 static int num_preferred_base;
  62 static struct progress *progress_state;
  63 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
  64 static int pack_compression_seen;
  65
  66 static struct packed_git *reuse_packfile;
  67 static uint32_t reuse_packfile_objects;
  68 static off_t reuse_packfile_offset;
  69
  70 static int use_bitmap_index_default = 1;
  71 static int use_bitmap_index = -1;
  72 static int write_bitmap_index;
  73 static uint16_t write_bitmap_options;
  74
  75 static unsigned long delta_cache_size = 0;
  76 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
  77 static unsigned long cache_max_small_delta_size = 1000;
  78
  79 static unsigned long window_memory_limit = 0;
  80
  81 /*
  82  * stats
  83  */
  84 static uint32_t written, written_delta;
  85 static uint32_t reused, reused_delta;
  86
  87 /*
  88  * Indexed commits
  89  */
  90 static struct commit **indexed_commits;
  91 static unsigned int indexed_commits_nr;
  92 static unsigned int indexed_commits_alloc;
  93
  94 static void index_commit_for_bitmap(struct commit *commit)
  95 {
  96         if (indexed_commits_nr >= indexed_commits_alloc) {
  97                 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
  98                 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
  99         }
 100
 101         indexed_commits[indexed_commits_nr++] = commit;
 102 }
 103
 104 static void *get_delta(struct object_entry *entry)
 105 {
 106         unsigned long size, base_size, delta_size;
 107         void *buf, *base_buf, *delta_buf;
 108         enum object_type type;
 109
 110         buf = read_sha1_file(entry->idx.sha1, &type, &size);
 111         if (!buf)
 112                 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
 113         base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
 114         if (!base_buf)
 115                 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
 116         delta_buf = diff_delta(base_buf, base_size,
 117                                buf, size, &delta_size, 0);
 118         if (!delta_buf || delta_size != entry->delta_size)
 119                 die("delta size changed");
 120         free(buf);
 121         free(base_buf);
 122         return delta_buf;
 123 }
 124
 125 static unsigned long do_compress(void **pptr, unsigned long size)
 126 {
 127         git_zstream stream;
 128         void *in, *out;
 129         unsigned long maxsize;
 130
 131         git_deflate_init(&stream, pack_compression_level);
 132         maxsize = git_deflate_bound(&stream, size);
 133
 134         in = *pptr;
 135         out = xmalloc(maxsize);
 136         *pptr = out;
 137
 138         stream.next_in = in;
 139         stream.avail_in = size;
 140         stream.next_out = out;
 141         stream.avail_out = maxsize;
 142         while (git_deflate(&stream, Z_FINISH) == Z_OK)
 143                 ; /* nothing */
 144         git_deflate_end(&stream);
 145
 146         free(in);
 147         return stream.total_out;
 148 }
 149
 150 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
 151                                            const unsigned char *sha1)
 152 {
 153         git_zstream stream;
 154         unsigned char ibuf[1024 * 16];
 155         unsigned char obuf[1024 * 16];
 156         unsigned long olen = 0;
 157
 158         git_deflate_init(&stream, pack_compression_level);
 159
 160         for (;;) {
 161                 ssize_t readlen;
 162                 int zret = Z_OK;
 163                 readlen = read_istream(st, ibuf, sizeof(ibuf));
 164                 if (readlen == -1)
 165                         die(_("unable to read %s"), sha1_to_hex(sha1));
 166
 167                 stream.next_in = ibuf;
 168                 stream.avail_in = readlen;
 169                 while ((stream.avail_in || readlen == 0) &&
 170                        (zret == Z_OK || zret == Z_BUF_ERROR)) {
 171                         stream.next_out = obuf;
 172                         stream.avail_out = sizeof(obuf);
 173                         zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 174                         sha1write(f, obuf, stream.next_out - obuf);
 175                         olen += stream.next_out - obuf;
 176                 }
 177                 if (stream.avail_in)
 178                         die(_("deflate error (%d)"), zret);
 179                 if (readlen == 0) {
 180                         if (zret != Z_STREAM_END)
 181                                 die(_("deflate error (%d)"), zret);
 182                         break;
 183                 }
 184         }
 185         git_deflate_end(&stream);
 186         return olen;
 187 }
 188
 189 /*
 190  * we are going to reuse the existing object data as is.  make
 191  * sure it is not corrupt.
 192  */
 193 static int check_pack_inflate(struct packed_git *p,
 194                 struct pack_window **w_curs,
 195                 off_t offset,
 196                 off_t len,
 197                 unsigned long expect)
 198 {
 199         git_zstream stream;
 200         unsigned char fakebuf[4096], *in;
 201         int st;
 202
 203         memset(&stream, 0, sizeof(stream));
 204         git_inflate_init(&stream);
 205         do {
 206                 in = use_pack(p, w_curs, offset, &stream.avail_in);
 207                 stream.next_in = in;
 208                 stream.next_out = fakebuf;
 209                 stream.avail_out = sizeof(fakebuf);
 210                 st = git_inflate(&stream, Z_FINISH);
 211                 offset += stream.next_in - in;
 212         } while (st == Z_OK || st == Z_BUF_ERROR);
 213         git_inflate_end(&stream);
 214         return (st == Z_STREAM_END &&
 215                 stream.total_out == expect &&
 216                 stream.total_in == len) ? 0 : -1;
 217 }
 218
 219 static void copy_pack_data(struct sha1file *f,
 220                 struct packed_git *p,
 221                 struct pack_window **w_curs,
 222                 off_t offset,
 223                 off_t len)
 224 {
 225         unsigned char *in;
 226         unsigned long avail;
 227
 228         while (len) {
 229                 in = use_pack(p, w_curs, offset, &avail);
 230                 if (avail > len)
 231                         avail = (unsigned long)len;
 232                 sha1write(f, in, avail);
 233                 offset += avail;
 234                 len -= avail;
 235         }
 236 }
 237
 238 /* Return 0 if we will bust the pack-size limit */
 239 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
 240                                            unsigned long limit, int usable_delta)
 241 {
 242         unsigned long size, datalen;
 243         unsigned char header[10], dheader[10];
 244         unsigned hdrlen;
 245         enum object_type type;
 246         void *buf;
 247         struct git_istream *st = NULL;
 248
 249         if (!usable_delta) {
 250                 if (entry->type == OBJ_BLOB &&
 251                     entry->size > big_file_threshold &&
 252                     (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
 253                         buf = NULL;
 254                 else {
 255                         buf = read_sha1_file(entry->idx.sha1, &type, &size);
 256                         if (!buf)
 257                                 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
 258                 }
 259                 /*
 260                  * make sure no cached delta data remains from a
 261                  * previous attempt before a pack split occurred.
 262                  */
 263                 free(entry->delta_data);
 264                 entry->delta_data = NULL;
 265                 entry->z_delta_size = 0;
 266         } else if (entry->delta_data) {
 267                 size = entry->delta_size;
 268                 buf = entry->delta_data;
 269                 entry->delta_data = NULL;
 270                 type = (allow_ofs_delta && entry->delta->idx.offset) ?
 271                         OBJ_OFS_DELTA : OBJ_REF_DELTA;
 272         } else {
 273                 buf = get_delta(entry);
 274                 size = entry->delta_size;
 275                 type = (allow_ofs_delta && entry->delta->idx.offset) ?
 276                         OBJ_OFS_DELTA : OBJ_REF_DELTA;
 277         }
 278
 279         if (st) /* large blob case, just assume we don't compress well */
 280                 datalen = size;
 281         else if (entry->z_delta_size)
 282                 datalen = entry->z_delta_size;
 283         else
 284                 datalen = do_compress(&buf, size);
 285
 286         /*
 287          * The object header is a byte of 'type' followed by zero or
 288          * more bytes of length.
 289          */
 290         hdrlen = encode_in_pack_object_header(type, size, header);
 291
 292         if (type == OBJ_OFS_DELTA) {
 293                 /*
 294                  * Deltas with relative base contain an additional
 295                  * encoding of the relative offset for the delta
 296                  * base from this object's position in the pack.
 297                  */
 298                 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 299                 unsigned pos = sizeof(dheader) - 1;
 300                 dheader[pos] = ofs & 127;
 301                 while (ofs >>= 7)
 302                         dheader[--pos] = 128 | (--ofs & 127);
 303                 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 304                         if (st)
 305                                 close_istream(st);
 306                         free(buf);
 307                         return 0;
 308                 }
 309                 sha1write(f, header, hdrlen);
 310                 sha1write(f, dheader + pos, sizeof(dheader) - pos);
 311                 hdrlen += sizeof(dheader) - pos;
 312         } else if (type == OBJ_REF_DELTA) {
 313                 /*
 314                  * Deltas with a base reference contain
 315                  * an additional 20 bytes for the base sha1.
 316                  */
 317                 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 318                         if (st)
 319                                 close_istream(st);
 320                         free(buf);
 321                         return 0;
 322                 }
 323                 sha1write(f, header, hdrlen);
 324                 sha1write(f, entry->delta->idx.sha1, 20);
 325                 hdrlen += 20;
 326         } else {
 327                 if (limit && hdrlen + datalen + 20 >= limit) {
 328                         if (st)
 329                                 close_istream(st);
 330                         free(buf);
 331                         return 0;
 332                 }
 333                 sha1write(f, header, hdrlen);
 334         }
 335         if (st) {
 336                 datalen = write_large_blob_data(st, f, entry->idx.sha1);
 337                 close_istream(st);
 338         } else {
 339                 sha1write(f, buf, datalen);
 340                 free(buf);
 341         }
 342
 343         return hdrlen + datalen;
 344 }
 345
 346 /* Return 0 if we will bust the pack-size limit */
 347 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
 348                                 unsigned long limit, int usable_delta)
 349 {
 350         struct packed_git *p = entry->in_pack;
 351         struct pack_window *w_curs = NULL;
 352         struct revindex_entry *revidx;
 353         off_t offset;
 354         enum object_type type = entry->type;
 355         off_t datalen;
 356         unsigned char header[10], dheader[10];
 357         unsigned hdrlen;
 358
 359         if (entry->delta)
 360                 type = (allow_ofs_delta && entry->delta->idx.offset) ?
 361                         OBJ_OFS_DELTA : OBJ_REF_DELTA;
 362         hdrlen = encode_in_pack_object_header(type, entry->size, header);
 363
 364         offset = entry->in_pack_offset;
 365         revidx = find_pack_revindex(p, offset);
 366         datalen = revidx[1].offset - offset;
 367         if (!pack_to_stdout && p->index_version > 1 &&
 368             check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 369                 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
 370                 unuse_pack(&w_curs);
 371                 return write_no_reuse_object(f, entry, limit, usable_delta);
 372         }
 373
 374         offset += entry->in_pack_header_size;
 375         datalen -= entry->in_pack_header_size;
 376
 377         if (!pack_to_stdout && p->index_version == 1 &&
 378             check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
 379                 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
 380                 unuse_pack(&w_curs);
 381                 return write_no_reuse_object(f, entry, limit, usable_delta);
 382         }
 383
 384         if (type == OBJ_OFS_DELTA) {
 385                 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 386                 unsigned pos = sizeof(dheader) - 1;
 387                 dheader[pos] = ofs & 127;
 388                 while (ofs >>= 7)
 389                         dheader[--pos] = 128 | (--ofs & 127);
 390                 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 391                         unuse_pack(&w_curs);
 392                         return 0;
 393                 }
 394                 sha1write(f, header, hdrlen);
 395                 sha1write(f, dheader + pos, sizeof(dheader) - pos);
 396                 hdrlen += sizeof(dheader) - pos;
 397                 reused_delta++;
 398         } else if (type == OBJ_REF_DELTA) {
 399                 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 400                         unuse_pack(&w_curs);
 401                         return 0;
 402                 }
 403                 sha1write(f, header, hdrlen);
 404                 sha1write(f, entry->delta->idx.sha1, 20);
 405                 hdrlen += 20;
 406                 reused_delta++;
 407         } else {
 408                 if (limit && hdrlen + datalen + 20 >= limit) {
 409                         unuse_pack(&w_curs);
 410                         return 0;
 411                 }
 412                 sha1write(f, header, hdrlen);
 413         }
 414         copy_pack_data(f, p, &w_curs, offset, datalen);
 415         unuse_pack(&w_curs);
 416         reused++;
 417         return hdrlen + datalen;
 418 }
 419
 420 /* Return 0 if we will bust the pack-size limit */
 421 static off_t write_object(struct sha1file *f,
 422                           struct object_entry *entry,
 423                           off_t write_offset)
 424 {
 425         unsigned long limit;
 426         off_t len;
 427         int usable_delta, to_reuse;
 428
 429         if (!pack_to_stdout)
 430                 crc32_begin(f);
 431
 432         /* apply size limit if limited packsize and not first object */
 433         if (!pack_size_limit || !nr_written)
 434                 limit = 0;
 435         else if (pack_size_limit <= write_offset)
 436                 /*
 437                  * the earlier object did not fit the limit; avoid
 438                  * mistaking this with unlimited (i.e. limit = 0).
 439                  */
 440                 limit = 1;
 441         else
 442                 limit = pack_size_limit - write_offset;
 443
 444         if (!entry->delta)
 445                 usable_delta = 0;       /* no delta */
 446         else if (!pack_size_limit)
 447                usable_delta = 1;        /* unlimited packfile */
 448         else if (entry->delta->idx.offset == (off_t)-1)
 449                 usable_delta = 0;       /* base was written to another pack */
 450         else if (entry->delta->idx.offset)
 451                 usable_delta = 1;       /* base already exists in this pack */
 452         else
 453                 usable_delta = 0;       /* base could end up in another pack */
 454
 455         if (!reuse_object)
 456                 to_reuse = 0;   /* explicit */
 457         else if (!entry->in_pack)
 458                 to_reuse = 0;   /* can't reuse what we don't have */
 459         else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
 460                                 /* check_object() decided it for us ... */
 461                 to_reuse = usable_delta;
 462                                 /* ... but pack split may override that */
 463         else if (entry->type != entry->in_pack_type)
 464                 to_reuse = 0;   /* pack has delta which is unusable */
 465         else if (entry->delta)
 466                 to_reuse = 0;   /* we want to pack afresh */
 467         else
 468                 to_reuse = 1;   /* we have it in-pack undeltified,
 469                                  * and we do not need to deltify it.
 470                                  */
 471
 472         if (!to_reuse)
 473                 len = write_no_reuse_object(f, entry, limit, usable_delta);
 474         else
 475                 len = write_reuse_object(f, entry, limit, usable_delta);
 476         if (!len)
 477                 return 0;
 478
 479         if (usable_delta)
 480                 written_delta++;
 481         written++;
 482         if (!pack_to_stdout)
 483                 entry->idx.crc32 = crc32_end(f);
 484         return len;
 485 }
 486
 487 enum write_one_status {
 488         WRITE_ONE_SKIP = -1, /* already written */
 489         WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 490         WRITE_ONE_WRITTEN = 1, /* normal */
 491         WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 492 };
 493
 494 static enum write_one_status write_one(struct sha1file *f,
 495                                        struct object_entry *e,
 496                                        off_t *offset)
 497 {
 498         off_t size;
 499         int recursing;
 500
 501         /*
 502          * we set offset to 1 (which is an impossible value) to mark
 503          * the fact that this object is involved in "write its base
 504          * first before writing a deltified object" recursion.
 505          */
 506         recursing = (e->idx.offset == 1);
 507         if (recursing) {
 508                 warning("recursive delta detected for object %s",
 509                         sha1_to_hex(e->idx.sha1));
 510                 return WRITE_ONE_RECURSIVE;
 511         } else if (e->idx.offset || e->preferred_base) {
 512                 /* offset is non zero if object is written already. */
 513                 return WRITE_ONE_SKIP;
 514         }
 515
 516         /* if we are deltified, write out base object first. */
 517         if (e->delta) {
 518                 e->idx.offset = 1; /* now recurse */
 519                 switch (write_one(f, e->delta, offset)) {
 520                 case WRITE_ONE_RECURSIVE:
 521                         /* we cannot depend on this one */
 522                         e->delta = NULL;
 523                         break;
 524                 default:
 525                         break;
 526                 case WRITE_ONE_BREAK:
 527                         e->idx.offset = recursing;
 528                         return WRITE_ONE_BREAK;
 529                 }
 530         }
 531
 532         e->idx.offset = *offset;
 533         size = write_object(f, e, *offset);
 534         if (!size) {
 535                 e->idx.offset = recursing;
 536                 return WRITE_ONE_BREAK;
 537         }
 538         written_list[nr_written++] = &e->idx;
 539
 540         /* make sure off_t is sufficiently large not to wrap */
 541         if (signed_add_overflows(*offset, size))
 542                 die("pack too large for current definition of off_t");
 543         *offset += size;
 544         return WRITE_ONE_WRITTEN;
 545 }
 546
 547 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 548                        void *cb_data)
 549 {
 550         unsigned char peeled[20];
 551         struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 552
 553         if (entry)
 554                 entry->tagged = 1;
 555         if (!peel_ref(path, peeled)) {
 556                 entry = packlist_find(&to_pack, peeled, NULL);
 557                 if (entry)
 558                         entry->tagged = 1;
 559         }
 560         return 0;
 561 }
 562
 563 static inline void add_to_write_order(struct object_entry **wo,
 564                                unsigned int *endp,
 565                                struct object_entry *e)
 566 {
 567         if (e->filled)
 568                 return;
 569         wo[(*endp)++] = e;
 570         e->filled = 1;
 571 }
 572
 573 static void add_descendants_to_write_order(struct object_entry **wo,
 574                                            unsigned int *endp,
 575                                            struct object_entry *e)
 576 {
 577         int add_to_order = 1;
 578         while (e) {
 579                 if (add_to_order) {
 580                         struct object_entry *s;
 581                         /* add this node... */
 582                         add_to_write_order(wo, endp, e);
 583                         /* all its siblings... */
 584                         for (s = e->delta_sibling; s; s = s->delta_sibling) {
 585                                 add_to_write_order(wo, endp, s);
 586                         }
 587                 }
 588                 /* drop down a level to add left subtree nodes if possible */
 589                 if (e->delta_child) {
 590                         add_to_order = 1;
 591                         e = e->delta_child;
 592                 } else {
 593                         add_to_order = 0;
 594                         /* our sibling might have some children, it is next */
 595                         if (e->delta_sibling) {
 596                                 e = e->delta_sibling;
 597                                 continue;
 598                         }
 599                         /* go back to our parent node */
 600                         e = e->delta;
 601                         while (e && !e->delta_sibling) {
 602                                 /* we're on the right side of a subtree, keep
 603                                  * going up until we can go right again */
 604                                 e = e->delta;
 605                         }
 606                         if (!e) {
 607                                 /* done- we hit our original root node */
 608                                 return;
 609                         }
 610                         /* pass it off to sibling at this level */
 611                         e = e->delta_sibling;
 612                 }
 613         };
 614 }
 615
 616 static void add_family_to_write_order(struct object_entry **wo,
 617                                       unsigned int *endp,
 618                                       struct object_entry *e)
 619 {
 620         struct object_entry *root;
 621
 622         for (root = e; root->delta; root = root->delta)
 623                 ; /* nothing */
 624         add_descendants_to_write_order(wo, endp, root);
 625 }
 626
 627 static struct object_entry **compute_write_order(void)
 628 {
 629         unsigned int i, wo_end, last_untagged;
 630
 631         struct object_entry **wo;
 632         struct object_entry *objects = to_pack.objects;
 633
 634         for (i = 0; i < to_pack.nr_objects; i++) {
 635                 objects[i].tagged = 0;
 636                 objects[i].filled = 0;
 637                 objects[i].delta_child = NULL;
 638                 objects[i].delta_sibling = NULL;
 639         }
 640
 641         /*
 642          * Fully connect delta_child/delta_sibling network.
 643          * Make sure delta_sibling is sorted in the original
 644          * recency order.
 645          */
 646         for (i = to_pack.nr_objects; i > 0;) {
 647                 struct object_entry *e = &objects[--i];
 648                 if (!e->delta)
 649                         continue;
 650                 /* Mark me as the first child */
 651                 e->delta_sibling = e->delta->delta_child;
 652                 e->delta->delta_child = e;
 653         }
 654
 655         /*
 656          * Mark objects that are at the tip of tags.
 657          */
 658         for_each_tag_ref(mark_tagged, NULL);
 659
 660         /*
 661          * Give the objects in the original recency order until
 662          * we see a tagged tip.
 663          */
 664         ALLOC_ARRAY(wo, to_pack.nr_objects);
 665         for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
 666                 if (objects[i].tagged)
 667                         break;
 668                 add_to_write_order(wo, &wo_end, &objects[i]);
 669         }
 670         last_untagged = i;
 671
 672         /*
 673          * Then fill all the tagged tips.
 674          */
 675         for (; i < to_pack.nr_objects; i++) {
 676                 if (objects[i].tagged)
 677                         add_to_write_order(wo, &wo_end, &objects[i]);
 678         }
 679
 680         /*
 681          * And then all remaining commits and tags.
 682          */
 683         for (i = last_untagged; i < to_pack.nr_objects; i++) {
 684                 if (objects[i].type != OBJ_COMMIT &&
 685                     objects[i].type != OBJ_TAG)
 686                         continue;
 687                 add_to_write_order(wo, &wo_end, &objects[i]);
 688         }
 689
 690         /*
 691          * And then all the trees.
 692          */
 693         for (i = last_untagged; i < to_pack.nr_objects; i++) {
 694                 if (objects[i].type != OBJ_TREE)
 695                         continue;
 696                 add_to_write_order(wo, &wo_end, &objects[i]);
 697         }
 698
 699         /*
 700          * Finally all the rest in really tight order
 701          */
 702         for (i = last_untagged; i < to_pack.nr_objects; i++) {
 703                 if (!objects[i].filled)
 704                         add_family_to_write_order(wo, &wo_end, &objects[i]);
 705         }
 706
 707         if (wo_end != to_pack.nr_objects)
 708                 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
 709
 710         return wo;
 711 }
 712
 713 static off_t write_reused_pack(struct sha1file *f)
 714 {
 715         unsigned char buffer[8192];
 716         off_t to_write, total;
 717         int fd;
 718
 719         if (!is_pack_valid(reuse_packfile))
 720                 die("packfile is invalid: %s", reuse_packfile->pack_name);
 721
 722         fd = git_open_noatime(reuse_packfile->pack_name);
 723         if (fd < 0)
 724                 die_errno("unable to open packfile for reuse: %s",
 725                           reuse_packfile->pack_name);
 726
 727         if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 728                 die_errno("unable to seek in reused packfile");
 729
 730         if (reuse_packfile_offset < 0)
 731                 reuse_packfile_offset = reuse_packfile->pack_size - 20;
 732
 733         total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 734
 735         while (to_write) {
 736                 int read_pack = xread(fd, buffer, sizeof(buffer));
 737
 738                 if (read_pack <= 0)
 739                         die_errno("unable to read from reused packfile");
 740
 741                 if (read_pack > to_write)
 742                         read_pack = to_write;
 743
 744                 sha1write(f, buffer, read_pack);
 745                 to_write -= read_pack;
 746
 747                 /*
 748                  * We don't know the actual number of objects written,
 749                  * only how many bytes written, how many bytes total, and
 750                  * how many objects total. So we can fake it by pretending all
 751                  * objects we are writing are the same size. This gives us a
 752                  * smooth progress meter, and at the end it matches the true
 753                  * answer.
 754                  */
 755                 written = reuse_packfile_objects *
 756                                 (((double)(total - to_write)) / total);
 757                 display_progress(progress_state, written);
 758         }
 759
 760         close(fd);
 761         written = reuse_packfile_objects;
 762         display_progress(progress_state, written);
 763         return reuse_packfile_offset - sizeof(struct pack_header);
 764 }
 765
 766 static const char no_split_warning[] = N_(
 767 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
 768 );
 769
 770 static void write_pack_file(void)
 771 {
 772         uint32_t i = 0, j;
 773         struct sha1file *f;
 774         off_t offset;
 775         uint32_t nr_remaining = nr_result;
 776         time_t last_mtime = 0;
 777         struct object_entry **write_order;
 778
 779         if (progress > pack_to_stdout)
 780                 progress_state = start_progress(_("Writing objects"), nr_result);
 781         ALLOC_ARRAY(written_list, to_pack.nr_objects);
 782         write_order = compute_write_order();
 783
 784         do {
 785                 unsigned char sha1[20];
 786                 char *pack_tmp_name = NULL;
 787
 788                 if (pack_to_stdout)
 789                         f = sha1fd_throughput(1, "<stdout>", progress_state);
 790                 else
 791                         f = create_tmp_packfile(&pack_tmp_name);
 792
 793                 offset = write_pack_header(f, nr_remaining);
 794
 795                 if (reuse_packfile) {
 796                         off_t packfile_size;
 797                         assert(pack_to_stdout);
 798
 799                         packfile_size = write_reused_pack(f);
 800                         offset += packfile_size;
 801                 }
 802
 803                 nr_written = 0;
 804                 for (; i < to_pack.nr_objects; i++) {
 805                         struct object_entry *e = write_order[i];
 806                         if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 807                                 break;
 808                         display_progress(progress_state, written);
 809                 }
 810
 811                 /*
 812                  * Did we write the wrong # entries in the header?
 813                  * If so, rewrite it like in fast-import
 814                  */
 815                 if (pack_to_stdout) {
 816                         sha1close(f, sha1, CSUM_CLOSE);
 817                 } else if (nr_written == nr_remaining) {
 818                         sha1close(f, sha1, CSUM_FSYNC);
 819                 } else {
 820                         int fd = sha1close(f, sha1, 0);
 821                         fixup_pack_header_footer(fd, sha1, pack_tmp_name,
 822                                                  nr_written, sha1, offset);
 823                         close(fd);
 824                         if (write_bitmap_index) {
 825                                 warning(_(no_split_warning));
 826                                 write_bitmap_index = 0;
 827                         }
 828                 }
 829
 830                 if (!pack_to_stdout) {
 831                         struct stat st;
 832                         struct strbuf tmpname = STRBUF_INIT;
 833
 834                         /*
 835                          * Packs are runtime accessed in their mtime
 836                          * order since newer packs are more likely to contain
 837                          * younger objects.  So if we are creating multiple
 838                          * packs then we should modify the mtime of later ones
 839                          * to preserve this property.
 840                          */
 841                         if (stat(pack_tmp_name, &st) < 0) {
 842                                 warning_errno("failed to stat %s", pack_tmp_name);
 843                         } else if (!last_mtime) {
 844                                 last_mtime = st.st_mtime;
 845                         } else {
 846                                 struct utimbuf utb;
 847                                 utb.actime = st.st_atime;
 848                                 utb.modtime = --last_mtime;
 849                                 if (utime(pack_tmp_name, &utb) < 0)
 850                                         warning_errno("failed utime() on %s", pack_tmp_name);
 851                         }
 852
 853                         strbuf_addf(&tmpname, "%s-", base_name);
 854
 855                         if (write_bitmap_index) {
 856                                 bitmap_writer_set_checksum(sha1);
 857                                 bitmap_writer_build_type_index(written_list, nr_written);
 858                         }
 859
 860                         finish_tmp_packfile(&tmpname, pack_tmp_name,
 861                                             written_list, nr_written,
 862                                             &pack_idx_opts, sha1);
 863
 864                         if (write_bitmap_index) {
 865                                 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
 866
 867                                 stop_progress(&progress_state);
 868
 869                                 bitmap_writer_show_progress(progress);
 870                                 bitmap_writer_reuse_bitmaps(&to_pack);
 871                                 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 872                                 bitmap_writer_build(&to_pack);
 873                                 bitmap_writer_finish(written_list, nr_written,
 874                                                      tmpname.buf, write_bitmap_options);
 875                                 write_bitmap_index = 0;
 876                         }
 877
 878                         strbuf_release(&tmpname);
 879                         free(pack_tmp_name);
 880                         puts(sha1_to_hex(sha1));
 881                 }
 882
 883                 /* mark written objects as written to previous pack */
 884                 for (j = 0; j < nr_written; j++) {
 885                         written_list[j]->offset = (off_t)-1;
 886                 }
 887                 nr_remaining -= nr_written;
 888         } while (nr_remaining && i < to_pack.nr_objects);
 889
 890         free(written_list);
 891         free(write_order);
 892         stop_progress(&progress_state);
 893         if (written != nr_result)
 894                 die("wrote %"PRIu32" objects while expecting %"PRIu32,
 895                         written, nr_result);
 896 }
 897
 898 static void setup_delta_attr_check(struct git_attr_check *check)
 899 {
 900         static struct git_attr *attr_delta;
 901
 902         if (!attr_delta)
 903                 attr_delta = git_attr("delta");
 904
 905         check[0].attr = attr_delta;
 906 }
 907
 908 static int no_try_delta(const char *path)
 909 {
 910         struct git_attr_check check[1];
 911
 912         setup_delta_attr_check(check);
 913         if (git_check_attr(path, ARRAY_SIZE(check), check))
 914                 return 0;
 915         if (ATTR_FALSE(check->value))
 916                 return 1;
 917         return 0;
 918 }
 919
 920 /*
 921  * When adding an object, check whether we have already added it
 922  * to our packing list. If so, we can skip. However, if we are
 923  * being asked to excludei t, but the previous mention was to include
 924  * it, make sure to adjust its flags and tweak our numbers accordingly.
 925  *
 926  * As an optimization, we pass out the index position where we would have
 927  * found the item, since that saves us from having to look it up again a
 928  * few lines later when we want to add the new entry.
 929  */
 930 static int have_duplicate_entry(const unsigned char *sha1,
 931                                 int exclude,
 932                                 uint32_t *index_pos)
 933 {
 934         struct object_entry *entry;
 935
 936         entry = packlist_find(&to_pack, sha1, index_pos);
 937         if (!entry)
 938                 return 0;
 939
 940         if (exclude) {
 941                 if (!entry->preferred_base)
 942                         nr_result--;
 943                 entry->preferred_base = 1;
 944         }
 945
 946         return 1;
 947 }
 948
 949 static int want_found_object(int exclude, struct packed_git *p)
 950 {
 951         if (exclude)
 952                 return 1;
 953         if (incremental)
 954                 return 0;
 955
 956         /*
 957          * When asked to do --local (do not include an object that appears in a
 958          * pack we borrow from elsewhere) or --honor-pack-keep (do not include
 959          * an object that appears in a pack marked with .keep), finding a pack
 960          * that matches the criteria is sufficient for us to decide to omit it.
 961          * However, even if this pack does not satisfy the criteria, we need to
 962          * make sure no copy of this object appears in _any_ pack that makes us
 963          * to omit the object, so we need to check all the packs.
 964          *
 965          * We can however first check whether these options can possible matter;
 966          * if they do not matter we know we want the object in generated pack.
 967          * Otherwise, we signal "-1" at the end to tell the caller that we do
 968          * not know either way, and it needs to check more packs.
 969          */
 970         if (!ignore_packed_keep &&
 971             (!local || !have_non_local_packs))
 972                 return 1;
 973
 974         if (local && !p->pack_local)
 975                 return 0;
 976         if (ignore_packed_keep && p->pack_local && p->pack_keep)
 977                 return 0;
 978
 979         /* we don't know yet; keep looking for more packs */
 980         return -1;
 981 }
 982
 983 /*
 984  * Check whether we want the object in the pack (e.g., we do not want
 985  * objects found in non-local stores if the "--local" option was used).
 986  *
 987  * If the caller already knows an existing pack it wants to take the object
 988  * from, that is passed in *found_pack and *found_offset; otherwise this
 989  * function finds if there is any pack that has the object and returns the pack
 990  * and its offset in these variables.
 991  */
 992 static int want_object_in_pack(const unsigned char *sha1,
 993                                int exclude,
 994                                struct packed_git **found_pack,
 995                                off_t *found_offset)
 996 {
 997         struct packed_git *p;
 998         int want;
 999
1000         if (!exclude && local && has_loose_object_nonlocal(sha1))
1001                 return 0;
1002
1003         /*
1004          * If we already know the pack object lives in, start checks from that
1005          * pack - in the usual case when neither --local was given nor .keep files
1006          * are present we will determine the answer right now.
1007          */
1008         if (*found_pack) {
1009                 want = want_found_object(exclude, *found_pack);
1010                 if (want != -1)
1011                         return want;
1012         }
1013
1014         for (p = packed_git; p; p = p->next) {
1015                 off_t offset;
1016
1017                 if (p == *found_pack)
1018                         offset = *found_offset;
1019                 else
1020                         offset = find_pack_entry_one(sha1, p);
1021
1022                 if (offset) {
1023                         if (!*found_pack) {
1024                                 if (!is_pack_valid(p))
1025                                         continue;
1026                                 *found_offset = offset;
1027                                 *found_pack = p;
1028                         }
1029                         want = want_found_object(exclude, p);
1030                         if (want != -1)
1031                                 return want;
1032                 }
1033         }
1034
1035         return 1;
1036 }
1037
1038 static void create_object_entry(const unsigned char *sha1,
1039                                 enum object_type type,
1040                                 uint32_t hash,
1041                                 int exclude,
1042                                 int no_try_delta,
1043                                 uint32_t index_pos,
1044                                 struct packed_git *found_pack,
1045                                 off_t found_offset)
1046 {
1047         struct object_entry *entry;
1048
1049         entry = packlist_alloc(&to_pack, sha1, index_pos);
1050         entry->hash = hash;
1051         if (type)
1052                 entry->type = type;
1053         if (exclude)
1054                 entry->preferred_base = 1;
1055         else
1056                 nr_result++;
1057         if (found_pack) {
1058                 entry->in_pack = found_pack;
1059                 entry->in_pack_offset = found_offset;
1060         }
1061
1062         entry->no_try_delta = no_try_delta;
1063 }
1064
1065 static const char no_closure_warning[] = N_(
1066 "disabling bitmap writing, as some objects are not being packed"
1067 );
1068
1069 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1070                             const char *name, int exclude)
1071 {
1072         struct packed_git *found_pack = NULL;
1073         off_t found_offset = 0;
1074         uint32_t index_pos;
1075
1076         if (have_duplicate_entry(sha1, exclude, &index_pos))
1077                 return 0;
1078
1079         if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1080                 /* The pack is missing an object, so it will not have closure */
1081                 if (write_bitmap_index) {
1082                         warning(_(no_closure_warning));
1083                         write_bitmap_index = 0;
1084                 }
1085                 return 0;
1086         }
1087
1088         create_object_entry(sha1, type, pack_name_hash(name),
1089                             exclude, name && no_try_delta(name),
1090                             index_pos, found_pack, found_offset);
1091
1092         display_progress(progress_state, nr_result);
1093         return 1;
1094 }
1095
1096 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1097                                         enum object_type type,
1098                                         int flags, uint32_t name_hash,
1099                                         struct packed_git *pack, off_t offset)
1100 {
1101         uint32_t index_pos;
1102
1103         if (have_duplicate_entry(sha1, 0, &index_pos))
1104                 return 0;
1105
1106         if (!want_object_in_pack(sha1, 0, &pack, &offset))
1107                 return 0;
1108
1109         create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1110
1111         display_progress(progress_state, nr_result);
1112         return 1;
1113 }
1114
1115 struct pbase_tree_cache {
1116         unsigned char sha1[20];
1117         int ref;
1118         int temporary;
1119         void *tree_data;
1120         unsigned long tree_size;
1121 };
1122
1123 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1124 static int pbase_tree_cache_ix(const unsigned char *sha1)
1125 {
1126         return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1127 }
1128 static int pbase_tree_cache_ix_incr(int ix)
1129 {
1130         return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1131 }
1132
1133 static struct pbase_tree {
1134         struct pbase_tree *next;
1135         /* This is a phony "cache" entry; we are not
1136          * going to evict it or find it through _get()
1137          * mechanism -- this is for the toplevel node that
1138          * would almost always change with any commit.
1139          */
1140         struct pbase_tree_cache pcache;
1141 } *pbase_tree;
1142
1143 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1144 {
1145         struct pbase_tree_cache *ent, *nent;
1146         void *data;
1147         unsigned long size;
1148         enum object_type type;
1149         int neigh;
1150         int my_ix = pbase_tree_cache_ix(sha1);
1151         int available_ix = -1;
1152
1153         /* pbase-tree-cache acts as a limited hashtable.
1154          * your object will be found at your index or within a few
1155          * slots after that slot if it is cached.
1156          */
1157         for (neigh = 0; neigh < 8; neigh++) {
1158                 ent = pbase_tree_cache[my_ix];
1159                 if (ent && !hashcmp(ent->sha1, sha1)) {
1160                         ent->ref++;
1161                         return ent;
1162                 }
1163                 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1164                          ((0 <= available_ix) &&
1165                           (!ent && pbase_tree_cache[available_ix])))
1166                         available_ix = my_ix;
1167                 if (!ent)
1168                         break;
1169                 my_ix = pbase_tree_cache_ix_incr(my_ix);
1170         }
1171
1172         /* Did not find one.  Either we got a bogus request or
1173          * we need to read and perhaps cache.
1174          */
1175         data = read_sha1_file(sha1, &type, &size);
1176         if (!data)
1177                 return NULL;
1178         if (type != OBJ_TREE) {
1179                 free(data);
1180                 return NULL;
1181         }
1182
1183         /* We need to either cache or return a throwaway copy */
1184
1185         if (available_ix < 0)
1186                 ent = NULL;
1187         else {
1188                 ent = pbase_tree_cache[available_ix];
1189                 my_ix = available_ix;
1190         }
1191
1192         if (!ent) {
1193                 nent = xmalloc(sizeof(*nent));
1194                 nent->temporary = (available_ix < 0);
1195         }
1196         else {
1197                 /* evict and reuse */
1198                 free(ent->tree_data);
1199                 nent = ent;
1200         }
1201         hashcpy(nent->sha1, sha1);
1202         nent->tree_data = data;
1203         nent->tree_size = size;
1204         nent->ref = 1;
1205         if (!nent->temporary)
1206                 pbase_tree_cache[my_ix] = nent;
1207         return nent;
1208 }
1209
1210 static void pbase_tree_put(struct pbase_tree_cache *cache)
1211 {
1212         if (!cache->temporary) {
1213                 cache->ref--;
1214                 return;
1215         }
1216         free(cache->tree_data);
1217         free(cache);
1218 }
1219
1220 static int name_cmp_len(const char *name)
1221 {
1222         int i;
1223         for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1224                 ;
1225         return i;
1226 }
1227
1228 static void add_pbase_object(struct tree_desc *tree,
1229                              const char *name,
1230                              int cmplen,
1231                              const char *fullname)
1232 {
1233         struct name_entry entry;
1234         int cmp;
1235
1236         while (tree_entry(tree,&entry)) {
1237                 if (S_ISGITLINK(entry.mode))
1238                         continue;
1239                 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1240                       memcmp(name, entry.path, cmplen);
1241                 if (cmp > 0)
1242                         continue;
1243                 if (cmp < 0)
1244                         return;
1245                 if (name[cmplen] != '/') {
1246                         add_object_entry(entry.oid->hash,
1247                                          object_type(entry.mode),
1248                                          fullname, 1);
1249                         return;
1250                 }
1251                 if (S_ISDIR(entry.mode)) {
1252                         struct tree_desc sub;
1253                         struct pbase_tree_cache *tree;
1254                         const char *down = name+cmplen+1;
1255                         int downlen = name_cmp_len(down);
1256
1257                         tree = pbase_tree_get(entry.oid->hash);
1258                         if (!tree)
1259                                 return;
1260                         init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1261
1262                         add_pbase_object(&sub, down, downlen, fullname);
1263                         pbase_tree_put(tree);
1264                 }
1265         }
1266 }
1267
1268 static unsigned *done_pbase_paths;
1269 static int done_pbase_paths_num;
1270 static int done_pbase_paths_alloc;
1271 static int done_pbase_path_pos(unsigned hash)
1272 {
1273         int lo = 0;
1274         int hi = done_pbase_paths_num;
1275         while (lo < hi) {
1276                 int mi = (hi + lo) / 2;
1277                 if (done_pbase_paths[mi] == hash)
1278                         return mi;
1279                 if (done_pbase_paths[mi] < hash)
1280                         hi = mi;
1281                 else
1282                         lo = mi + 1;
1283         }
1284         return -lo-1;
1285 }
1286
1287 static int check_pbase_path(unsigned hash)
1288 {
1289         int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1290         if (0 <= pos)
1291                 return 1;
1292         pos = -pos - 1;
1293         ALLOC_GROW(done_pbase_paths,
1294                    done_pbase_paths_num + 1,
1295                    done_pbase_paths_alloc);
1296         done_pbase_paths_num++;
1297         if (pos < done_pbase_paths_num)
1298                 memmove(done_pbase_paths + pos + 1,
1299                         done_pbase_paths + pos,
1300                         (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1301         done_pbase_paths[pos] = hash;
1302         return 0;
1303 }
1304
1305 static void add_preferred_base_object(const char *name)
1306 {
1307         struct pbase_tree *it;
1308         int cmplen;
1309         unsigned hash = pack_name_hash(name);
1310
1311         if (!num_preferred_base || check_pbase_path(hash))
1312                 return;
1313
1314         cmplen = name_cmp_len(name);
1315         for (it = pbase_tree; it; it = it->next) {
1316                 if (cmplen == 0) {
1317                         add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1318                 }
1319                 else {
1320                         struct tree_desc tree;
1321                         init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1322                         add_pbase_object(&tree, name, cmplen, name);
1323                 }
1324         }
1325 }
1326
1327 static void add_preferred_base(unsigned char *sha1)
1328 {
1329         struct pbase_tree *it;
1330         void *data;
1331         unsigned long size;
1332         unsigned char tree_sha1[20];
1333
1334         if (window <= num_preferred_base++)
1335                 return;
1336
1337         data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1338         if (!data)
1339                 return;
1340
1341         for (it = pbase_tree; it; it = it->next) {
1342                 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1343                         free(data);
1344                         return;
1345                 }
1346         }
1347
1348         it = xcalloc(1, sizeof(*it));
1349         it->next = pbase_tree;
1350         pbase_tree = it;
1351
1352         hashcpy(it->pcache.sha1, tree_sha1);
1353         it->pcache.tree_data = data;
1354         it->pcache.tree_size = size;
1355 }
1356
1357 static void cleanup_preferred_base(void)
1358 {
1359         struct pbase_tree *it;
1360         unsigned i;
1361
1362         it = pbase_tree;
1363         pbase_tree = NULL;
1364         while (it) {
1365                 struct pbase_tree *this = it;
1366                 it = this->next;
1367                 free(this->pcache.tree_data);
1368                 free(this);
1369         }
1370
1371         for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1372                 if (!pbase_tree_cache[i])
1373                         continue;
1374                 free(pbase_tree_cache[i]->tree_data);
1375                 free(pbase_tree_cache[i]);
1376                 pbase_tree_cache[i] = NULL;
1377         }
1378
1379         free(done_pbase_paths);
1380         done_pbase_paths = NULL;
1381         done_pbase_paths_num = done_pbase_paths_alloc = 0;
1382 }
1383
1384 static void check_object(struct object_entry *entry)
1385 {
1386         if (entry->in_pack) {
1387                 struct packed_git *p = entry->in_pack;
1388                 struct pack_window *w_curs = NULL;
1389                 const unsigned char *base_ref = NULL;
1390                 struct object_entry *base_entry;
1391                 unsigned long used, used_0;
1392                 unsigned long avail;
1393                 off_t ofs;
1394                 unsigned char *buf, c;
1395
1396                 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1397
1398                 /*
1399                  * We want in_pack_type even if we do not reuse delta
1400                  * since non-delta representations could still be reused.
1401                  */
1402                 used = unpack_object_header_buffer(buf, avail,
1403                                                    &entry->in_pack_type,
1404                                                    &entry->size);
1405                 if (used == 0)
1406                         goto give_up;
1407
1408                 /*
1409                  * Determine if this is a delta and if so whether we can
1410                  * reuse it or not.  Otherwise let's find out as cheaply as
1411                  * possible what the actual type and size for this object is.
1412                  */
1413                 switch (entry->in_pack_type) {
1414                 default:
1415                         /* Not a delta hence we've already got all we need. */
1416                         entry->type = entry->in_pack_type;
1417                         entry->in_pack_header_size = used;
1418                         if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1419                                 goto give_up;
1420                         unuse_pack(&w_curs);
1421                         return;
1422                 case OBJ_REF_DELTA:
1423                         if (reuse_delta && !entry->preferred_base)
1424                                 base_ref = use_pack(p, &w_curs,
1425                                                 entry->in_pack_offset + used, NULL);
1426                         entry->in_pack_header_size = used + 20;
1427                         break;
1428                 case OBJ_OFS_DELTA:
1429                         buf = use_pack(p, &w_curs,
1430                                        entry->in_pack_offset + used, NULL);
1431                         used_0 = 0;
1432                         c = buf[used_0++];
1433                         ofs = c & 127;
1434                         while (c & 128) {
1435                                 ofs += 1;
1436                                 if (!ofs || MSB(ofs, 7)) {
1437                                         error("delta base offset overflow in pack for %s",
1438                                               sha1_to_hex(entry->idx.sha1));
1439                                         goto give_up;
1440                                 }
1441                                 c = buf[used_0++];
1442                                 ofs = (ofs << 7) + (c & 127);
1443                         }
1444                         ofs = entry->in_pack_offset - ofs;
1445                         if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1446                                 error("delta base offset out of bound for %s",
1447                                       sha1_to_hex(entry->idx.sha1));
1448                                 goto give_up;
1449                         }
1450                         if (reuse_delta && !entry->preferred_base) {
1451                                 struct revindex_entry *revidx;
1452                                 revidx = find_pack_revindex(p, ofs);
1453                                 if (!revidx)
1454                                         goto give_up;
1455                                 base_ref = nth_packed_object_sha1(p, revidx->nr);
1456                         }
1457                         entry->in_pack_header_size = used + used_0;
1458                         break;
1459                 }
1460
1461                 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1462                         /*
1463                          * If base_ref was set above that means we wish to
1464                          * reuse delta data, and we even found that base
1465                          * in the list of objects we want to pack. Goodie!
1466                          *
1467                          * Depth value does not matter - find_deltas() will
1468                          * never consider reused delta as the base object to
1469                          * deltify other objects against, in order to avoid
1470                          * circular deltas.
1471                          */
1472                         entry->type = entry->in_pack_type;
1473                         entry->delta = base_entry;
1474                         entry->delta_size = entry->size;
1475                         entry->delta_sibling = base_entry->delta_child;
1476                         base_entry->delta_child = entry;
1477                         unuse_pack(&w_curs);
1478                         return;
1479                 }
1480
1481                 if (entry->type) {
1482                         /*
1483                          * This must be a delta and we already know what the
1484                          * final object type is.  Let's extract the actual
1485                          * object size from the delta header.
1486                          */
1487                         entry->size = get_size_from_delta(p, &w_curs,
1488                                         entry->in_pack_offset + entry->in_pack_header_size);
1489                         if (entry->size == 0)
1490                                 goto give_up;
1491                         unuse_pack(&w_curs);
1492                         return;
1493                 }
1494
1495                 /*
1496                  * No choice but to fall back to the recursive delta walk
1497                  * with sha1_object_info() to find about the object type
1498                  * at this point...
1499                  */
1500                 give_up:
1501                 unuse_pack(&w_curs);
1502         }
1503
1504         entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1505         /*
1506          * The error condition is checked in prepare_pack().  This is
1507          * to permit a missing preferred base object to be ignored
1508          * as a preferred base.  Doing so can result in a larger
1509          * pack file, but the transfer will still take place.
1510          */
1511 }
1512
1513 static int pack_offset_sort(const void *_a, const void *_b)
1514 {
1515         const struct object_entry *a = *(struct object_entry **)_a;
1516         const struct object_entry *b = *(struct object_entry **)_b;
1517
1518         /* avoid filesystem trashing with loose objects */
1519         if (!a->in_pack && !b->in_pack)
1520                 return hashcmp(a->idx.sha1, b->idx.sha1);
1521
1522         if (a->in_pack < b->in_pack)
1523                 return -1;
1524         if (a->in_pack > b->in_pack)
1525                 return 1;
1526         return a->in_pack_offset < b->in_pack_offset ? -1 :
1527                         (a->in_pack_offset > b->in_pack_offset);
1528 }
1529
1530 static void get_object_details(void)
1531 {
1532         uint32_t i;
1533         struct object_entry **sorted_by_offset;
1534
1535         sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1536         for (i = 0; i < to_pack.nr_objects; i++)
1537                 sorted_by_offset[i] = to_pack.objects + i;
1538         qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1539
1540         for (i = 0; i < to_pack.nr_objects; i++) {
1541                 struct object_entry *entry = sorted_by_offset[i];
1542                 check_object(entry);
1543                 if (big_file_threshold < entry->size)
1544                         entry->no_try_delta = 1;
1545         }
1546
1547         free(sorted_by_offset);
1548 }
1549
1550 /*
1551  * We search for deltas in a list sorted by type, by filename hash, and then
1552  * by size, so that we see progressively smaller and smaller files.
1553  * That's because we prefer deltas to be from the bigger file
1554  * to the smaller -- deletes are potentially cheaper, but perhaps
1555  * more importantly, the bigger file is likely the more recent
1556  * one.  The deepest deltas are therefore the oldest objects which are
1557  * less susceptible to be accessed often.
1558  */
1559 static int type_size_sort(const void *_a, const void *_b)
1560 {
1561         const struct object_entry *a = *(struct object_entry **)_a;
1562         const struct object_entry *b = *(struct object_entry **)_b;
1563
1564         if (a->type > b->type)
1565                 return -1;
1566         if (a->type < b->type)
1567                 return 1;
1568         if (a->hash > b->hash)
1569                 return -1;
1570         if (a->hash < b->hash)
1571                 return 1;
1572         if (a->preferred_base > b->preferred_base)
1573                 return -1;
1574         if (a->preferred_base < b->preferred_base)
1575                 return 1;
1576         if (a->size > b->size)
1577                 return -1;
1578         if (a->size < b->size)
1579                 return 1;
1580         return a < b ? -1 : (a > b);  /* newest first */
1581 }
1582
1583 struct unpacked {
1584         struct object_entry *entry;
1585         void *data;
1586         struct delta_index *index;
1587         unsigned depth;
1588 };
1589
1590 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1591                            unsigned long delta_size)
1592 {
1593         if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1594                 return 0;
1595
1596         if (delta_size < cache_max_small_delta_size)
1597                 return 1;
1598
1599         /* cache delta, if objects are large enough compared to delta size */
1600         if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1601                 return 1;
1602
1603         return 0;
1604 }
1605
1606 #ifndef NO_PTHREADS
1607
1608 static pthread_mutex_t read_mutex;
1609 #define read_lock()             pthread_mutex_lock(&read_mutex)
1610 #define read_unlock()           pthread_mutex_unlock(&read_mutex)
1611
1612 static pthread_mutex_t cache_mutex;
1613 #define cache_lock()            pthread_mutex_lock(&cache_mutex)
1614 #define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1615
1616 static pthread_mutex_t progress_mutex;
1617 #define progress_lock()         pthread_mutex_lock(&progress_mutex)
1618 #define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1619
1620 #else
1621
1622 #define read_lock()             (void)0
1623 #define read_unlock()           (void)0
1624 #define cache_lock()            (void)0
1625 #define cache_unlock()          (void)0
1626 #define progress_lock()         (void)0
1627 #define progress_unlock()       (void)0
1628
1629 #endif
1630
1631 static int try_delta(struct unpacked *trg, struct unpacked *src,
1632                      unsigned max_depth, unsigned long *mem_usage)
1633 {
1634         struct object_entry *trg_entry = trg->entry;
1635         struct object_entry *src_entry = src->entry;
1636         unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1637         unsigned ref_depth;
1638         enum object_type type;
1639         void *delta_buf;
1640
1641         /* Don't bother doing diffs between different types */
1642         if (trg_entry->type != src_entry->type)
1643                 return -1;
1644
1645         /*
1646          * We do not bother to try a delta that we discarded on an
1647          * earlier try, but only when reusing delta data.  Note that
1648          * src_entry that is marked as the preferred_base should always
1649          * be considered, as even if we produce a suboptimal delta against
1650          * it, we will still save the transfer cost, as we already know
1651          * the other side has it and we won't send src_entry at all.
1652          */
1653         if (reuse_delta && trg_entry->in_pack &&
1654             trg_entry->in_pack == src_entry->in_pack &&
1655             !src_entry->preferred_base &&
1656             trg_entry->in_pack_type != OBJ_REF_DELTA &&
1657             trg_entry->in_pack_type != OBJ_OFS_DELTA)
1658                 return 0;
1659
1660         /* Let's not bust the allowed depth. */
1661         if (src->depth >= max_depth)
1662                 return 0;
1663
1664         /* Now some size filtering heuristics. */
1665         trg_size = trg_entry->size;
1666         if (!trg_entry->delta) {
1667                 max_size = trg_size/2 - 20;
1668                 ref_depth = 1;
1669         } else {
1670                 max_size = trg_entry->delta_size;
1671                 ref_depth = trg->depth;
1672         }
1673         max_size = (uint64_t)max_size * (max_depth - src->depth) /
1674                                                 (max_depth - ref_depth + 1);
1675         if (max_size == 0)
1676                 return 0;
1677         src_size = src_entry->size;
1678         sizediff = src_size < trg_size ? trg_size - src_size : 0;
1679         if (sizediff >= max_size)
1680                 return 0;
1681         if (trg_size < src_size / 32)
1682                 return 0;
1683
1684         /* Load data if not already done */
1685         if (!trg->data) {
1686                 read_lock();
1687                 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1688                 read_unlock();
1689                 if (!trg->data)
1690                         die("object %s cannot be read",
1691                             sha1_to_hex(trg_entry->idx.sha1));
1692                 if (sz != trg_size)
1693                         die("object %s inconsistent object length (%lu vs %lu)",
1694                             sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1695                 *mem_usage += sz;
1696         }
1697         if (!src->data) {
1698                 read_lock();
1699                 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1700                 read_unlock();
1701                 if (!src->data) {
1702                         if (src_entry->preferred_base) {
1703                                 static int warned = 0;
1704                                 if (!warned++)
1705                                         warning("object %s cannot be read",
1706                                                 sha1_to_hex(src_entry->idx.sha1));
1707                                 /*
1708                                  * Those objects are not included in the
1709                                  * resulting pack.  Be resilient and ignore
1710                                  * them if they can't be read, in case the
1711                                  * pack could be created nevertheless.
1712                                  */
1713                                 return 0;
1714                         }
1715                         die("object %s cannot be read",
1716                             sha1_to_hex(src_entry->idx.sha1));
1717                 }
1718                 if (sz != src_size)
1719                         die("object %s inconsistent object length (%lu vs %lu)",
1720                             sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1721                 *mem_usage += sz;
1722         }
1723         if (!src->index) {
1724                 src->index = create_delta_index(src->data, src_size);
1725                 if (!src->index) {
1726                         static int warned = 0;
1727                         if (!warned++)
1728                                 warning("suboptimal pack - out of memory");
1729                         return 0;
1730                 }
1731                 *mem_usage += sizeof_delta_index(src->index);
1732         }
1733
1734         delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1735         if (!delta_buf)
1736                 return 0;
1737
1738         if (trg_entry->delta) {
1739                 /* Prefer only shallower same-sized deltas. */
1740                 if (delta_size == trg_entry->delta_size &&
1741                     src->depth + 1 >= trg->depth) {
1742                         free(delta_buf);
1743                         return 0;
1744                 }
1745         }
1746
1747         /*
1748          * Handle memory allocation outside of the cache
1749          * accounting lock.  Compiler will optimize the strangeness
1750          * away when NO_PTHREADS is defined.
1751          */
1752         free(trg_entry->delta_data);
1753         cache_lock();
1754         if (trg_entry->delta_data) {
1755                 delta_cache_size -= trg_entry->delta_size;
1756                 trg_entry->delta_data = NULL;
1757         }
1758         if (delta_cacheable(src_size, trg_size, delta_size)) {
1759                 delta_cache_size += delta_size;
1760                 cache_unlock();
1761                 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1762         } else {
1763                 cache_unlock();
1764                 free(delta_buf);
1765         }
1766
1767         trg_entry->delta = src_entry;
1768         trg_entry->delta_size = delta_size;
1769         trg->depth = src->depth + 1;
1770
1771         return 1;
1772 }
1773
1774 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1775 {
1776         struct object_entry *child = me->delta_child;
1777         unsigned int m = n;
1778         while (child) {
1779                 unsigned int c = check_delta_limit(child, n + 1);
1780                 if (m < c)
1781                         m = c;
1782                 child = child->delta_sibling;
1783         }
1784         return m;
1785 }
1786
1787 static unsigned long free_unpacked(struct unpacked *n)
1788 {
1789         unsigned long freed_mem = sizeof_delta_index(n->index);
1790         free_delta_index(n->index);
1791         n->index = NULL;
1792         if (n->data) {
1793                 freed_mem += n->entry->size;
1794                 free(n->data);
1795                 n->data = NULL;
1796         }
1797         n->entry = NULL;
1798         n->depth = 0;
1799         return freed_mem;
1800 }
1801
1802 static void find_deltas(struct object_entry **list, unsigned *list_size,
1803                         int window, int depth, unsigned *processed)
1804 {
1805         uint32_t i, idx = 0, count = 0;
1806         struct unpacked *array;
1807         unsigned long mem_usage = 0;
1808
1809         array = xcalloc(window, sizeof(struct unpacked));
1810
1811         for (;;) {
1812                 struct object_entry *entry;
1813                 struct unpacked *n = array + idx;
1814                 int j, max_depth, best_base = -1;
1815
1816                 progress_lock();
1817                 if (!*list_size) {
1818                         progress_unlock();
1819                         break;
1820                 }
1821                 entry = *list++;
1822                 (*list_size)--;
1823                 if (!entry->preferred_base) {
1824                         (*processed)++;
1825                         display_progress(progress_state, *processed);
1826                 }
1827                 progress_unlock();
1828
1829                 mem_usage -= free_unpacked(n);
1830                 n->entry = entry;
1831
1832                 while (window_memory_limit &&
1833                        mem_usage > window_memory_limit &&
1834                        count > 1) {
1835                         uint32_t tail = (idx + window - count) % window;
1836                         mem_usage -= free_unpacked(array + tail);
1837                         count--;
1838                 }
1839
1840                 /* We do not compute delta to *create* objects we are not
1841                  * going to pack.
1842                  */
1843                 if (entry->preferred_base)
1844                         goto next;
1845
1846                 /*
1847                  * If the current object is at pack edge, take the depth the
1848                  * objects that depend on the current object into account
1849                  * otherwise they would become too deep.
1850                  */
1851                 max_depth = depth;
1852                 if (entry->delta_child) {
1853                         max_depth -= check_delta_limit(entry, 0);
1854                         if (max_depth <= 0)
1855                                 goto next;
1856                 }
1857
1858                 j = window;
1859                 while (--j > 0) {
1860                         int ret;
1861                         uint32_t other_idx = idx + j;
1862                         struct unpacked *m;
1863                         if (other_idx >= window)
1864                                 other_idx -= window;
1865                         m = array + other_idx;
1866                         if (!m->entry)
1867                                 break;
1868                         ret = try_delta(n, m, max_depth, &mem_usage);
1869                         if (ret < 0)
1870                                 break;
1871                         else if (ret > 0)
1872                                 best_base = other_idx;
1873                 }
1874
1875                 /*
1876                  * If we decided to cache the delta data, then it is best
1877                  * to compress it right away.  First because we have to do
1878                  * it anyway, and doing it here while we're threaded will
1879                  * save a lot of time in the non threaded write phase,
1880                  * as well as allow for caching more deltas within
1881                  * the same cache size limit.
1882                  * ...
1883                  * But only if not writing to stdout, since in that case
1884                  * the network is most likely throttling writes anyway,
1885                  * and therefore it is best to go to the write phase ASAP
1886                  * instead, as we can afford spending more time compressing
1887                  * between writes at that moment.
1888                  */
1889                 if (entry->delta_data && !pack_to_stdout) {
1890                         entry->z_delta_size = do_compress(&entry->delta_data,
1891                                                           entry->delta_size);
1892                         cache_lock();
1893                         delta_cache_size -= entry->delta_size;
1894                         delta_cache_size += entry->z_delta_size;
1895                         cache_unlock();
1896                 }
1897
1898                 /* if we made n a delta, and if n is already at max
1899                  * depth, leaving it in the window is pointless.  we
1900                  * should evict it first.
1901                  */
1902                 if (entry->delta && max_depth <= n->depth)
1903                         continue;
1904
1905                 /*
1906                  * Move the best delta base up in the window, after the
1907                  * currently deltified object, to keep it longer.  It will
1908                  * be the first base object to be attempted next.
1909                  */
1910                 if (entry->delta) {
1911                         struct unpacked swap = array[best_base];
1912                         int dist = (window + idx - best_base) % window;
1913                         int dst = best_base;
1914                         while (dist--) {
1915                                 int src = (dst + 1) % window;
1916                                 array[dst] = array[src];
1917                                 dst = src;
1918                         }
1919                         array[dst] = swap;
1920                 }
1921
1922                 next:
1923                 idx++;
1924                 if (count + 1 < window)
1925                         count++;
1926                 if (idx >= window)
1927                         idx = 0;
1928         }
1929
1930         for (i = 0; i < window; ++i) {
1931                 free_delta_index(array[i].index);
1932                 free(array[i].data);
1933         }
1934         free(array);
1935 }
1936
1937 #ifndef NO_PTHREADS
1938
1939 static void try_to_free_from_threads(size_t size)
1940 {
1941         read_lock();
1942         release_pack_memory(size);
1943         read_unlock();
1944 }
1945
1946 static try_to_free_t old_try_to_free_routine;
1947
1948 /*
1949  * The main thread waits on the condition that (at least) one of the workers
1950  * has stopped working (which is indicated in the .working member of
1951  * struct thread_params).
1952  * When a work thread has completed its work, it sets .working to 0 and
1953  * signals the main thread and waits on the condition that .data_ready
1954  * becomes 1.
1955  */
1956
1957 struct thread_params {
1958         pthread_t thread;
1959         struct object_entry **list;
1960         unsigned list_size;
1961         unsigned remaining;
1962         int window;
1963         int depth;
1964         int working;
1965         int data_ready;
1966         pthread_mutex_t mutex;
1967         pthread_cond_t cond;
1968         unsigned *processed;
1969 };
1970
1971 static pthread_cond_t progress_cond;
1972
1973 /*
1974  * Mutex and conditional variable can't be statically-initialized on Windows.
1975  */
1976 static void init_threaded_search(void)
1977 {
1978         init_recursive_mutex(&read_mutex);
1979         pthread_mutex_init(&cache_mutex, NULL);
1980         pthread_mutex_init(&progress_mutex, NULL);
1981         pthread_cond_init(&progress_cond, NULL);
1982         old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1983 }
1984
1985 static void cleanup_threaded_search(void)
1986 {
1987         set_try_to_free_routine(old_try_to_free_routine);
1988         pthread_cond_destroy(&progress_cond);
1989         pthread_mutex_destroy(&read_mutex);
1990         pthread_mutex_destroy(&cache_mutex);
1991         pthread_mutex_destroy(&progress_mutex);
1992 }
1993
1994 static void *threaded_find_deltas(void *arg)
1995 {
1996         struct thread_params *me = arg;
1997
1998         while (me->remaining) {
1999                 find_deltas(me->list, &me->remaining,
2000                             me->window, me->depth, me->processed);
2001
2002                 progress_lock();
2003                 me->working = 0;
2004                 pthread_cond_signal(&progress_cond);
2005                 progress_unlock();
2006
2007                 /*
2008                  * We must not set ->data_ready before we wait on the
2009                  * condition because the main thread may have set it to 1
2010                  * before we get here. In order to be sure that new
2011                  * work is available if we see 1 in ->data_ready, it
2012                  * was initialized to 0 before this thread was spawned
2013                  * and we reset it to 0 right away.
2014                  */
2015                 pthread_mutex_lock(&me->mutex);
2016                 while (!me->data_ready)
2017                         pthread_cond_wait(&me->cond, &me->mutex);
2018                 me->data_ready = 0;
2019                 pthread_mutex_unlock(&me->mutex);
2020         }
2021         /* leave ->working 1 so that this doesn't get more work assigned */
2022         return NULL;
2023 }
2024
2025 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2026                            int window, int depth, unsigned *processed)
2027 {
2028         struct thread_params *p;
2029         int i, ret, active_threads = 0;
2030
2031         init_threaded_search();
2032
2033         if (delta_search_threads <= 1) {
2034                 find_deltas(list, &list_size, window, depth, processed);
2035                 cleanup_threaded_search();
2036                 return;
2037         }
2038         if (progress > pack_to_stdout)
2039                 fprintf(stderr, "Delta compression using up to %d threads.\n",
2040                                 delta_search_threads);
2041         p = xcalloc(delta_search_threads, sizeof(*p));
2042
2043         /* Partition the work amongst work threads. */
2044         for (i = 0; i < delta_search_threads; i++) {
2045                 unsigned sub_size = list_size / (delta_search_threads - i);
2046
2047                 /* don't use too small segments or no deltas will be found */
2048                 if (sub_size < 2*window && i+1 < delta_search_threads)
2049                         sub_size = 0;
2050
2051                 p[i].window = window;
2052                 p[i].depth = depth;
2053                 p[i].processed = processed;
2054                 p[i].working = 1;
2055                 p[i].data_ready = 0;
2056
2057                 /* try to split chunks on "path" boundaries */
2058                 while (sub_size && sub_size < list_size &&
2059                        list[sub_size]->hash &&
2060                        list[sub_size]->hash == list[sub_size-1]->hash)
2061                         sub_size++;
2062
2063                 p[i].list = list;
2064                 p[i].list_size = sub_size;
2065                 p[i].remaining = sub_size;
2066
2067                 list += sub_size;
2068                 list_size -= sub_size;
2069         }
2070
2071         /* Start work threads. */
2072         for (i = 0; i < delta_search_threads; i++) {
2073                 if (!p[i].list_size)
2074                         continue;
2075                 pthread_mutex_init(&p[i].mutex, NULL);
2076                 pthread_cond_init(&p[i].cond, NULL);
2077                 ret = pthread_create(&p[i].thread, NULL,
2078                                      threaded_find_deltas, &p[i]);
2079                 if (ret)
2080                         die("unable to create thread: %s", strerror(ret));
2081                 active_threads++;
2082         }
2083
2084         /*
2085          * Now let's wait for work completion.  Each time a thread is done
2086          * with its work, we steal half of the remaining work from the
2087          * thread with the largest number of unprocessed objects and give
2088          * it to that newly idle thread.  This ensure good load balancing
2089          * until the remaining object list segments are simply too short
2090          * to be worth splitting anymore.
2091          */
2092         while (active_threads) {
2093                 struct thread_params *target = NULL;
2094                 struct thread_params *victim = NULL;
2095                 unsigned sub_size = 0;
2096
2097                 progress_lock();
2098                 for (;;) {
2099                         for (i = 0; !target && i < delta_search_threads; i++)
2100                                 if (!p[i].working)
2101                                         target = &p[i];
2102                         if (target)
2103                                 break;
2104                         pthread_cond_wait(&progress_cond, &progress_mutex);
2105                 }
2106
2107                 for (i = 0; i < delta_search_threads; i++)
2108                         if (p[i].remaining > 2*window &&
2109                             (!victim || victim->remaining < p[i].remaining))
2110                                 victim = &p[i];
2111                 if (victim) {
2112                         sub_size = victim->remaining / 2;
2113                         list = victim->list + victim->list_size - sub_size;
2114                         while (sub_size && list[0]->hash &&
2115                                list[0]->hash == list[-1]->hash) {
2116                                 list++;
2117                                 sub_size--;
2118                         }
2119                         if (!sub_size) {
2120                                 /*
2121                                  * It is possible for some "paths" to have
2122                                  * so many objects that no hash boundary
2123                                  * might be found.  Let's just steal the
2124                                  * exact half in that case.
2125                                  */
2126                                 sub_size = victim->remaining / 2;
2127                                 list -= sub_size;
2128                         }
2129                         target->list = list;
2130                         victim->list_size -= sub_size;
2131                         victim->remaining -= sub_size;
2132                 }
2133                 target->list_size = sub_size;
2134                 target->remaining = sub_size;
2135                 target->working = 1;
2136                 progress_unlock();
2137
2138                 pthread_mutex_lock(&target->mutex);
2139                 target->data_ready = 1;
2140                 pthread_cond_signal(&target->cond);
2141                 pthread_mutex_unlock(&target->mutex);
2142
2143                 if (!sub_size) {
2144                         pthread_join(target->thread, NULL);
2145                         pthread_cond_destroy(&target->cond);
2146                         pthread_mutex_destroy(&target->mutex);
2147                         active_threads--;
2148                 }
2149         }
2150         cleanup_threaded_search();
2151         free(p);
2152 }
2153
2154 #else
2155 #define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2156 #endif
2157
2158 static void add_tag_chain(const struct object_id *oid)
2159 {
2160         struct tag *tag;
2161
2162         /*
2163          * We catch duplicates already in add_object_entry(), but we'd
2164          * prefer to do this extra check to avoid having to parse the
2165          * tag at all if we already know that it's being packed (e.g., if
2166          * it was included via bitmaps, we would not have parsed it
2167          * previously).
2168          */
2169         if (packlist_find(&to_pack, oid->hash, NULL))
2170                 return;
2171
2172         tag = lookup_tag(oid->hash);
2173         while (1) {
2174                 if (!tag || parse_tag(tag) || !tag->tagged)
2175                         die("unable to pack objects reachable from tag %s",
2176                             oid_to_hex(oid));
2177
2178                 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2179
2180                 if (tag->tagged->type != OBJ_TAG)
2181                         return;
2182
2183                 tag = (struct tag *)tag->tagged;
2184         }
2185 }
2186
2187 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2188 {
2189         struct object_id peeled;
2190
2191         if (starts_with(path, "refs/tags/") && /* is a tag? */
2192             !peel_ref(path, peeled.hash)    && /* peelable? */
2193             packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2194                 add_tag_chain(oid);
2195         return 0;
2196 }
2197
2198 static void prepare_pack(int window, int depth)
2199 {
2200         struct object_entry **delta_list;
2201         uint32_t i, nr_deltas;
2202         unsigned n;
2203
2204         get_object_details();
2205
2206         /*
2207          * If we're locally repacking then we need to be doubly careful
2208          * from now on in order to make sure no stealth corruption gets
2209          * propagated to the new pack.  Clients receiving streamed packs
2210          * should validate everything they get anyway so no need to incur
2211          * the additional cost here in that case.
2212          */
2213         if (!pack_to_stdout)
2214                 do_check_packed_object_crc = 1;
2215
2216         if (!to_pack.nr_objects || !window || !depth)
2217                 return;
2218
2219         ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2220         nr_deltas = n = 0;
2221
2222         for (i = 0; i < to_pack.nr_objects; i++) {
2223                 struct object_entry *entry = to_pack.objects + i;
2224
2225                 if (entry->delta)
2226                         /* This happens if we decided to reuse existing
2227                          * delta from a pack.  "reuse_delta &&" is implied.
2228                          */
2229                         continue;
2230
2231                 if (entry->size < 50)
2232                         continue;
2233
2234                 if (entry->no_try_delta)
2235                         continue;
2236
2237                 if (!entry->preferred_base) {
2238                         nr_deltas++;
2239                         if (entry->type < 0)
2240                                 die("unable to get type of object %s",
2241                                     sha1_to_hex(entry->idx.sha1));
2242                 } else {
2243                         if (entry->type < 0) {
2244                                 /*
2245                                  * This object is not found, but we
2246                                  * don't have to include it anyway.
2247                                  */
2248                                 continue;
2249                         }
2250                 }
2251
2252                 delta_list[n++] = entry;
2253         }
2254
2255         if (nr_deltas && n > 1) {
2256                 unsigned nr_done = 0;
2257                 if (progress)
2258                         progress_state = start_progress(_("Compressing objects"),
2259                                                         nr_deltas);
2260                 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2261                 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2262                 stop_progress(&progress_state);
2263                 if (nr_done != nr_deltas)
2264                         die("inconsistency with delta count");
2265         }
2266         free(delta_list);
2267 }
2268
2269 static int git_pack_config(const char *k, const char *v, void *cb)
2270 {
2271         if (!strcmp(k, "pack.window")) {
2272                 window = git_config_int(k, v);
2273                 return 0;
2274         }
2275         if (!strcmp(k, "pack.windowmemory")) {
2276                 window_memory_limit = git_config_ulong(k, v);
2277                 return 0;
2278         }
2279         if (!strcmp(k, "pack.depth")) {
2280                 depth = git_config_int(k, v);
2281                 return 0;
2282         }
2283         if (!strcmp(k, "pack.compression")) {
2284                 int level = git_config_int(k, v);
2285                 if (level == -1)
2286                         level = Z_DEFAULT_COMPRESSION;
2287                 else if (level < 0 || level > Z_BEST_COMPRESSION)
2288                         die("bad pack compression level %d", level);
2289                 pack_compression_level = level;
2290                 pack_compression_seen = 1;
2291                 return 0;
2292         }
2293         if (!strcmp(k, "pack.deltacachesize")) {
2294                 max_delta_cache_size = git_config_int(k, v);
2295                 return 0;
2296         }
2297         if (!strcmp(k, "pack.deltacachelimit")) {
2298                 cache_max_small_delta_size = git_config_int(k, v);
2299                 return 0;
2300         }
2301         if (!strcmp(k, "pack.writebitmaphashcache")) {
2302                 if (git_config_bool(k, v))
2303                         write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2304                 else
2305                         write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2306         }
2307         if (!strcmp(k, "pack.usebitmaps")) {
2308                 use_bitmap_index_default = git_config_bool(k, v);
2309                 return 0;
2310         }
2311         if (!strcmp(k, "pack.threads")) {
2312                 delta_search_threads = git_config_int(k, v);
2313                 if (delta_search_threads < 0)
2314                         die("invalid number of threads specified (%d)",
2315                             delta_search_threads);
2316 #ifdef NO_PTHREADS
2317                 if (delta_search_threads != 1)
2318                         warning("no threads support, ignoring %s", k);
2319 #endif
2320                 return 0;
2321         }
2322         if (!strcmp(k, "pack.indexversion")) {
2323                 pack_idx_opts.version = git_config_int(k, v);
2324                 if (pack_idx_opts.version > 2)
2325                         die("bad pack.indexversion=%"PRIu32,
2326                             pack_idx_opts.version);
2327                 return 0;
2328         }
2329         return git_default_config(k, v, cb);
2330 }
2331
2332 static void read_object_list_from_stdin(void)
2333 {
2334         char line[40 + 1 + PATH_MAX + 2];
2335         unsigned char sha1[20];
2336
2337         for (;;) {
2338                 if (!fgets(line, sizeof(line), stdin)) {
2339                         if (feof(stdin))
2340                                 break;
2341                         if (!ferror(stdin))
2342                                 die("fgets returned NULL, not EOF, not error!");
2343                         if (errno != EINTR)
2344                                 die_errno("fgets");
2345                         clearerr(stdin);
2346                         continue;
2347                 }
2348                 if (line[0] == '-') {
2349                         if (get_sha1_hex(line+1, sha1))
2350                                 die("expected edge sha1, got garbage:\n %s",
2351                                     line);
2352                         add_preferred_base(sha1);
2353                         continue;
2354                 }
2355                 if (get_sha1_hex(line, sha1))
2356                         die("expected sha1, got garbage:\n %s", line);
2357
2358                 add_preferred_base_object(line+41);
2359                 add_object_entry(sha1, 0, line+41, 0);
2360         }
2361 }
2362
2363 #define OBJECT_ADDED (1u<<20)
2364
2365 static void show_commit(struct commit *commit, void *data)
2366 {
2367         add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2368         commit->object.flags |= OBJECT_ADDED;
2369
2370         if (write_bitmap_index)
2371                 index_commit_for_bitmap(commit);
2372 }
2373
2374 static void show_object(struct object *obj, const char *name, void *data)
2375 {
2376         add_preferred_base_object(name);
2377         add_object_entry(obj->oid.hash, obj->type, name, 0);
2378         obj->flags |= OBJECT_ADDED;
2379 }
2380
2381 static void show_edge(struct commit *commit)
2382 {
2383         add_preferred_base(commit->object.oid.hash);
2384 }
2385
2386 struct in_pack_object {
2387         off_t offset;
2388         struct object *object;
2389 };
2390
2391 struct in_pack {
2392         int alloc;
2393         int nr;
2394         struct in_pack_object *array;
2395 };
2396
2397 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2398 {
2399         in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2400         in_pack->array[in_pack->nr].object = object;
2401         in_pack->nr++;
2402 }
2403
2404 /*
2405  * Compare the objects in the offset order, in order to emulate the
2406  * "git rev-list --objects" output that produced the pack originally.
2407  */
2408 static int ofscmp(const void *a_, const void *b_)
2409 {
2410         struct in_pack_object *a = (struct in_pack_object *)a_;
2411         struct in_pack_object *b = (struct in_pack_object *)b_;
2412
2413         if (a->offset < b->offset)
2414                 return -1;
2415         else if (a->offset > b->offset)
2416                 return 1;
2417         else
2418                 return oidcmp(&a->object->oid, &b->object->oid);
2419 }
2420
2421 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2422 {
2423         struct packed_git *p;
2424         struct in_pack in_pack;
2425         uint32_t i;
2426
2427         memset(&in_pack, 0, sizeof(in_pack));
2428
2429         for (p = packed_git; p; p = p->next) {
2430                 const unsigned char *sha1;
2431                 struct object *o;
2432
2433                 if (!p->pack_local || p->pack_keep)
2434                         continue;
2435                 if (open_pack_index(p))
2436                         die("cannot open pack index");
2437
2438                 ALLOC_GROW(in_pack.array,
2439                            in_pack.nr + p->num_objects,
2440                            in_pack.alloc);
2441
2442                 for (i = 0; i < p->num_objects; i++) {
2443                         sha1 = nth_packed_object_sha1(p, i);
2444                         o = lookup_unknown_object(sha1);
2445                         if (!(o->flags & OBJECT_ADDED))
2446                                 mark_in_pack_object(o, p, &in_pack);
2447                         o->flags |= OBJECT_ADDED;
2448                 }
2449         }
2450
2451         if (in_pack.nr) {
2452                 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2453                       ofscmp);
2454                 for (i = 0; i < in_pack.nr; i++) {
2455                         struct object *o = in_pack.array[i].object;
2456                         add_object_entry(o->oid.hash, o->type, "", 0);
2457                 }
2458         }
2459         free(in_pack.array);
2460 }
2461
2462 static int add_loose_object(const unsigned char *sha1, const char *path,
2463                             void *data)
2464 {
2465         enum object_type type = sha1_object_info(sha1, NULL);
2466
2467         if (type < 0) {
2468                 warning("loose object at %s could not be examined", path);
2469                 return 0;
2470         }
2471
2472         add_object_entry(sha1, type, "", 0);
2473         return 0;
2474 }
2475
2476 /*
2477  * We actually don't even have to worry about reachability here.
2478  * add_object_entry will weed out duplicates, so we just add every
2479  * loose object we find.
2480  */
2481 static void add_unreachable_loose_objects(void)
2482 {
2483         for_each_loose_file_in_objdir(get_object_directory(),
2484                                       add_loose_object,
2485                                       NULL, NULL, NULL);
2486 }
2487
2488 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2489 {
2490         static struct packed_git *last_found = (void *)1;
2491         struct packed_git *p;
2492
2493         p = (last_found != (void *)1) ? last_found : packed_git;
2494
2495         while (p) {
2496                 if ((!p->pack_local || p->pack_keep) &&
2497                         find_pack_entry_one(sha1, p)) {
2498                         last_found = p;
2499                         return 1;
2500                 }
2501                 if (p == last_found)
2502                         p = packed_git;
2503                 else
2504                         p = p->next;
2505                 if (p == last_found)
2506                         p = p->next;
2507         }
2508         return 0;
2509 }
2510
2511 /*
2512  * Store a list of sha1s that are should not be discarded
2513  * because they are either written too recently, or are
2514  * reachable from another object that was.
2515  *
2516  * This is filled by get_object_list.
2517  */
2518 static struct sha1_array recent_objects;
2519
2520 static int loosened_object_can_be_discarded(const unsigned char *sha1,
2521                                             unsigned long mtime)
2522 {
2523         if (!unpack_unreachable_expiration)
2524                 return 0;
2525         if (mtime > unpack_unreachable_expiration)
2526                 return 0;
2527         if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2528                 return 0;
2529         return 1;
2530 }
2531
2532 static void loosen_unused_packed_objects(struct rev_info *revs)
2533 {
2534         struct packed_git *p;
2535         uint32_t i;
2536         const unsigned char *sha1;
2537
2538         for (p = packed_git; p; p = p->next) {
2539                 if (!p->pack_local || p->pack_keep)
2540                         continue;
2541
2542                 if (open_pack_index(p))
2543                         die("cannot open pack index");
2544
2545                 for (i = 0; i < p->num_objects; i++) {
2546                         sha1 = nth_packed_object_sha1(p, i);
2547                         if (!packlist_find(&to_pack, sha1, NULL) &&
2548                             !has_sha1_pack_kept_or_nonlocal(sha1) &&
2549                             !loosened_object_can_be_discarded(sha1, p->mtime))
2550                                 if (force_object_loose(sha1, p->mtime))
2551                                         die("unable to force loose object");
2552                 }
2553         }
2554 }
2555
2556 /*
2557  * This tracks any options which pack-reuse code expects to be on, or which a
2558  * reader of the pack might not understand, and which would therefore prevent
2559  * blind reuse of what we have on disk.
2560  */
2561 static int pack_options_allow_reuse(void)
2562 {
2563         return pack_to_stdout && allow_ofs_delta;
2564 }
2565
2566 static int get_object_list_from_bitmap(struct rev_info *revs)
2567 {
2568         if (prepare_bitmap_walk(revs) < 0)
2569                 return -1;
2570
2571         if (pack_options_allow_reuse() &&
2572             !reuse_partial_packfile_from_bitmap(
2573                         &reuse_packfile,
2574                         &reuse_packfile_objects,
2575                         &reuse_packfile_offset)) {
2576                 assert(reuse_packfile_objects);
2577                 nr_result += reuse_packfile_objects;
2578                 display_progress(progress_state, nr_result);
2579         }
2580
2581         traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2582         return 0;
2583 }
2584
2585 static void record_recent_object(struct object *obj,
2586                                  const char *name,
2587                                  void *data)
2588 {
2589         sha1_array_append(&recent_objects, obj->oid.hash);
2590 }
2591
2592 static void record_recent_commit(struct commit *commit, void *data)
2593 {
2594         sha1_array_append(&recent_objects, commit->object.oid.hash);
2595 }
2596
2597 static void get_object_list(int ac, const char **av)
2598 {
2599         struct rev_info revs;
2600         char line[1000];
2601         int flags = 0;
2602
2603         init_revisions(&revs, NULL);
2604         save_commit_buffer = 0;
2605         setup_revisions(ac, av, &revs, NULL);
2606
2607         /* make sure shallows are read */
2608         is_repository_shallow();
2609
2610         while (fgets(line, sizeof(line), stdin) != NULL) {
2611                 int len = strlen(line);
2612                 if (len && line[len - 1] == '\n')
2613                         line[--len] = 0;
2614                 if (!len)
2615                         break;
2616                 if (*line == '-') {
2617                         if (!strcmp(line, "--not")) {
2618                                 flags ^= UNINTERESTING;
2619                                 write_bitmap_index = 0;
2620                                 continue;
2621                         }
2622                         if (starts_with(line, "--shallow ")) {
2623                                 unsigned char sha1[20];
2624                                 if (get_sha1_hex(line + 10, sha1))
2625                                         die("not an SHA-1 '%s'", line + 10);
2626                                 register_shallow(sha1);
2627                                 use_bitmap_index = 0;
2628                                 continue;
2629                         }
2630                         die("not a rev '%s'", line);
2631                 }
2632                 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2633                         die("bad revision '%s'", line);
2634         }
2635
2636         if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2637                 return;
2638
2639         if (prepare_revision_walk(&revs))
2640                 die("revision walk setup failed");
2641         mark_edges_uninteresting(&revs, show_edge);
2642         traverse_commit_list(&revs, show_commit, show_object, NULL);
2643
2644         if (unpack_unreachable_expiration) {
2645                 revs.ignore_missing_links = 1;
2646                 if (add_unseen_recent_objects_to_traversal(&revs,
2647                                 unpack_unreachable_expiration))
2648                         die("unable to add recent objects");
2649                 if (prepare_revision_walk(&revs))
2650                         die("revision walk setup failed");
2651                 traverse_commit_list(&revs, record_recent_commit,
2652                                      record_recent_object, NULL);
2653         }
2654
2655         if (keep_unreachable)
2656                 add_objects_in_unpacked_packs(&revs);
2657         if (pack_loose_unreachable)
2658                 add_unreachable_loose_objects();
2659         if (unpack_unreachable)
2660                 loosen_unused_packed_objects(&revs);
2661
2662         sha1_array_clear(&recent_objects);
2663 }
2664
2665 static int option_parse_index_version(const struct option *opt,
2666                                       const char *arg, int unset)
2667 {
2668         char *c;
2669         const char *val = arg;
2670         pack_idx_opts.version = strtoul(val, &c, 10);
2671         if (pack_idx_opts.version > 2)
2672                 die(_("unsupported index version %s"), val);
2673         if (*c == ',' && c[1])
2674                 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2675         if (*c || pack_idx_opts.off32_limit & 0x80000000)
2676                 die(_("bad index version '%s'"), val);
2677         return 0;
2678 }
2679
2680 static int option_parse_unpack_unreachable(const struct option *opt,
2681                                            const char *arg, int unset)
2682 {
2683         if (unset) {
2684                 unpack_unreachable = 0;
2685                 unpack_unreachable_expiration = 0;
2686         }
2687         else {
2688                 unpack_unreachable = 1;
2689                 if (arg)
2690                         unpack_unreachable_expiration = approxidate(arg);
2691         }
2692         return 0;
2693 }
2694
2695 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2696 {
2697         int use_internal_rev_list = 0;
2698         int thin = 0;
2699         int shallow = 0;
2700         int all_progress_implied = 0;
2701         struct argv_array rp = ARGV_ARRAY_INIT;
2702         int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2703         int rev_list_index = 0;
2704         struct option pack_objects_options[] = {
2705                 OPT_SET_INT('q', "quiet", &progress,
2706                             N_("do not show progress meter"), 0),
2707                 OPT_SET_INT(0, "progress", &progress,
2708                             N_("show progress meter"), 1),
2709                 OPT_SET_INT(0, "all-progress", &progress,
2710                             N_("show progress meter during object writing phase"), 2),
2711                 OPT_BOOL(0, "all-progress-implied",
2712                          &all_progress_implied,
2713                          N_("similar to --all-progress when progress meter is shown")),
2714                 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2715                   N_("write the pack index file in the specified idx format version"),
2716                   0, option_parse_index_version },
2717                 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2718                               N_("maximum size of each output pack file")),
2719                 OPT_BOOL(0, "local", &local,
2720                          N_("ignore borrowed objects from alternate object store")),
2721                 OPT_BOOL(0, "incremental", &incremental,
2722                          N_("ignore packed objects")),
2723                 OPT_INTEGER(0, "window", &window,
2724                             N_("limit pack window by objects")),
2725                 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2726                               N_("limit pack window by memory in addition to object limit")),
2727                 OPT_INTEGER(0, "depth", &depth,
2728                             N_("maximum length of delta chain allowed in the resulting pack")),
2729                 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2730                          N_("reuse existing deltas")),
2731                 OPT_BOOL(0, "reuse-object", &reuse_object,
2732                          N_("reuse existing objects")),
2733                 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2734                          N_("use OFS_DELTA objects")),
2735                 OPT_INTEGER(0, "threads", &delta_search_threads,
2736                             N_("use threads when searching for best delta matches")),
2737                 OPT_BOOL(0, "non-empty", &non_empty,
2738                          N_("do not create an empty pack output")),
2739                 OPT_BOOL(0, "revs", &use_internal_rev_list,
2740                          N_("read revision arguments from standard input")),
2741                 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2742                   N_("limit the objects to those that are not yet packed"),
2743                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2744                 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2745                   N_("include objects reachable from any reference"),
2746                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2747                 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2748                   N_("include objects referred by reflog entries"),
2749                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2750                 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2751                   N_("include objects referred to by the index"),
2752                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2753                 OPT_BOOL(0, "stdout", &pack_to_stdout,
2754                          N_("output pack to stdout")),
2755                 OPT_BOOL(0, "include-tag", &include_tag,
2756                          N_("include tag objects that refer to objects to be packed")),
2757                 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2758                          N_("keep unreachable objects")),
2759                 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2760                          N_("pack loose unreachable objects")),
2761                 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2762                   N_("unpack unreachable objects newer than <time>"),
2763                   PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2764                 OPT_BOOL(0, "thin", &thin,
2765                          N_("create thin packs")),
2766                 OPT_BOOL(0, "shallow", &shallow,
2767                          N_("create packs suitable for shallow fetches")),
2768                 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2769                          N_("ignore packs that have companion .keep file")),
2770                 OPT_INTEGER(0, "compression", &pack_compression_level,
2771                             N_("pack compression level")),
2772                 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2773                             N_("do not hide commits by grafts"), 0),
2774                 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2775                          N_("use a bitmap index if available to speed up counting objects")),
2776                 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2777                          N_("write a bitmap index together with the pack index")),
2778                 OPT_END(),
2779         };
2780
2781         check_replace_refs = 0;
2782
2783         reset_pack_idx_option(&pack_idx_opts);
2784         git_config(git_pack_config, NULL);
2785         if (!pack_compression_seen && core_compression_seen)
2786                 pack_compression_level = core_compression_level;
2787
2788         progress = isatty(2);
2789         argc = parse_options(argc, argv, prefix, pack_objects_options,
2790                              pack_usage, 0);
2791
2792         if (argc) {
2793                 base_name = argv[0];
2794                 argc--;
2795         }
2796         if (pack_to_stdout != !base_name || argc)
2797                 usage_with_options(pack_usage, pack_objects_options);
2798
2799         argv_array_push(&rp, "pack-objects");
2800         if (thin) {
2801                 use_internal_rev_list = 1;
2802                 argv_array_push(&rp, shallow
2803                                 ? "--objects-edge-aggressive"
2804                                 : "--objects-edge");
2805         } else
2806                 argv_array_push(&rp, "--objects");
2807
2808         if (rev_list_all) {
2809                 use_internal_rev_list = 1;
2810                 argv_array_push(&rp, "--all");
2811         }
2812         if (rev_list_reflog) {
2813                 use_internal_rev_list = 1;
2814                 argv_array_push(&rp, "--reflog");
2815         }
2816         if (rev_list_index) {
2817                 use_internal_rev_list = 1;
2818                 argv_array_push(&rp, "--indexed-objects");
2819         }
2820         if (rev_list_unpacked) {
2821                 use_internal_rev_list = 1;
2822                 argv_array_push(&rp, "--unpacked");
2823         }
2824
2825         if (!reuse_object)
2826                 reuse_delta = 0;
2827         if (pack_compression_level == -1)
2828                 pack_compression_level = Z_DEFAULT_COMPRESSION;
2829         else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2830                 die("bad pack compression level %d", pack_compression_level);
2831
2832         if (!delta_search_threads)      /* --threads=0 means autodetect */
2833                 delta_search_threads = online_cpus();
2834
2835 #ifdef NO_PTHREADS
2836         if (delta_search_threads != 1)
2837                 warning("no threads support, ignoring --threads");
2838 #endif
2839         if (!pack_to_stdout && !pack_size_limit)
2840                 pack_size_limit = pack_size_limit_cfg;
2841         if (pack_to_stdout && pack_size_limit)
2842                 die("--max-pack-size cannot be used to build a pack for transfer.");
2843         if (pack_size_limit && pack_size_limit < 1024*1024) {
2844                 warning("minimum pack size limit is 1 MiB");
2845                 pack_size_limit = 1024*1024;
2846         }
2847
2848         if (!pack_to_stdout && thin)
2849                 die("--thin cannot be used to build an indexable pack.");
2850
2851         if (keep_unreachable && unpack_unreachable)
2852                 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2853         if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2854                 unpack_unreachable_expiration = 0;
2855
2856         /*
2857          * "soft" reasons not to use bitmaps - for on-disk repack by default we want
2858          *
2859          * - to produce good pack (with bitmap index not-yet-packed objects are
2860          *   packed in suboptimal order).
2861          *
2862          * - to use more robust pack-generation codepath (avoiding possible
2863          *   bugs in bitmap code and possible bitmap index corruption).
2864          */
2865         if (!pack_to_stdout)
2866                 use_bitmap_index_default = 0;
2867
2868         if (use_bitmap_index < 0)
2869                 use_bitmap_index = use_bitmap_index_default;
2870
2871         /* "hard" reasons not to use bitmaps; these just won't work at all */
2872         if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
2873                 use_bitmap_index = 0;
2874
2875         if (pack_to_stdout || !rev_list_all)
2876                 write_bitmap_index = 0;
2877
2878         if (progress && all_progress_implied)
2879                 progress = 2;
2880
2881         prepare_packed_git();
2882         if (ignore_packed_keep) {
2883                 struct packed_git *p;
2884                 for (p = packed_git; p; p = p->next)
2885                         if (p->pack_local && p->pack_keep)
2886                                 break;
2887                 if (!p) /* no keep-able packs found */
2888                         ignore_packed_keep = 0;
2889         }
2890         if (local) {
2891                 /*
2892                  * unlike ignore_packed_keep above, we do not want to
2893                  * unset "local" based on looking at packs, as it
2894                  * also covers non-local objects
2895                  */
2896                 struct packed_git *p;
2897                 for (p = packed_git; p; p = p->next) {
2898                         if (!p->pack_local) {
2899                                 have_non_local_packs = 1;
2900                                 break;
2901                         }
2902                 }
2903         }
2904
2905         if (progress)
2906                 progress_state = start_progress(_("Counting objects"), 0);
2907         if (!use_internal_rev_list)
2908                 read_object_list_from_stdin();
2909         else {
2910                 get_object_list(rp.argc, rp.argv);
2911                 argv_array_clear(&rp);
2912         }
2913         cleanup_preferred_base();
2914         if (include_tag && nr_result)
2915                 for_each_ref(add_ref_tag, NULL);
2916         stop_progress(&progress_state);
2917
2918         if (non_empty && !nr_result)
2919                 return 0;
2920         if (nr_result)
2921                 prepare_pack(window, depth);
2922         write_pack_file();
2923         if (progress)
2924                 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2925                         " reused %"PRIu32" (delta %"PRIu32")\n",
2926                         written, written_delta, reused, reused_delta);
2927         return 0;
2928 }